Phys-JEPA: Physics-Informed Latent World Models for Multivariate Time-Series Forecasting

The article introduces Phys-JEPA, a physics-informed joint-embedding predictive architecture designed for multivariate time-series forecasting. This model decomposes predictive states into physical and residual components, imposing physical consistency directly on latent states and transitions, which enhances the model's interpretability and accuracy. Initial benchmarks on datasets such as Jena Climate and Traffic show significant reductions in mean squared error (MSE), indicating that integrating physics-informed learning within the latent space can improve forecasting performance in complex systems.